A Comparison of Current Graph Database Models

Renzo Angles Universidad de Talca (Chile)

3rd Int. Workshop on Graph Data Management: Techniques and applications (GDM 2012) 5 April, Washington DC, USA

Outline 1. Introduction 2. Current graph databases 3. Comparison of graph database models 4. Conclusions and future work

Outline 1. Introduction 2. Current graph databases 3. Comparison of graph database models 4. Conclusions and future work

Database models
A data model is a collection of conceptual tools used to model real-world entities and the relationships among them [Silberschatz et al. 1996]. A DB model consists of 3 components [Codd 1980]:

(2) Query operators

(3) Integrity rules

(1) Data structure types

Graph database model

1 Graph data structures
Simple graphs (nodes + edges + labels + direction) Generalizations: nested graphs (hypernodes), hypergraphs (hyperedges), attributed graphs Simple functions (e.g., shortest-path) Graph query language (operators) Domain-specific queries: graph pattern mining Schema-instance consistency Identification of nodes, atributes and relations Path constraints

2 Graph-oriented operations

3 Graph integrity constraints

Graph databases (before 2003)

1975 1984 ... 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1999 2002
R&M LDM G-Base O2 Tompa GOOD GROOVI GMOD PaMal GraphDB G-Log GRAS GOQL GDM W&X Gram GOAL DGV GGL Hypernode Simatic-XT Hy+ Hypernode2 Hypernode3

R. Angles and C. Gutierrez. Survey of Graph Database Models. ACM Comp. Surveys, 2008.

Graph databases (after 2003)

200 4 200 5 200 6 200 7 200 8 200 9 201 0 201 1 201 VertexDB Hypergraph CloudGraph Pregel InfiniteGraph Trinity Sones Filament GStore Horton AllegroGraph

DEX

Neo4j

Motivation
1. What is the most suitable graph database?
Empirical comparison: desirable but hard Benchmark: there is not a standard one The application domain is also important

2. Objective: comparison of graph data models

Independent of implementation Easier to evaluate (vs empirical evaluation) Shows (in advance) the expressive power for data modeling

Outline 1. Introduction 2. Current graph databases 3. Comparison of graph database models 4. Conclusions and future work

Current graph databases

Graph database systems AllegroGraph (2005): SW-oriented databases DEX (2007): bitmaps-based graph database Neo4J (2007): disk-based transactional graph database HyperGraphDB (2010): hypergraph-based database InfiniteGraph (2010): distributed-oriented system Sones (2010): object-oriented database

Current graph databases

Graph stores VertexDB (2009): key-value disk store (TokyoCabinet) Filament (2010): graph library on PostgreSQL G-Store (2010): prototype Redis_graph (2010): implemented on python Work in progress Pregel (Google, 2009): vertex-based intrastructure for graphs Horton (Microsoft, 2010): transactional graph processing CloudGraph (2010): use MySQL as backed Trinity (Microsoft, 2011): RAM-based key value store

Current graph databases

Related DB technologies Web-oriented DBs: InfoGrid , FlockDB Document-oriented DBs: OrientDB Triple stores (RDF DBs): 4Store, Virtuoso, Bigdata Distributed graph processing: Angrapa, Apache Hama, Giraph, GoldenOrb, Phoebus, KDT, Signal Collect, HipG

Outline 1. Introduction 2. Current graph databases 3. Comparison of graph database models 4. Conclusions and future work

Data storing features

Backed: RDB (filament), BerkeleyDB (HypergraphDB), TokyoCabinet (vertexDB) Indexing: nodes, relations, atributes, triples Data formats: GraphML, graphViz, N-Triples, RDF-XML, ACID (partial support): Allegro, HypergraphDB, Infinite, Neo4j

Operation and manipulation features

Languages: SPARQL 1.0(.1), GraphQL, (Sones) Why is an API the main approach? does it facilities the development of applications?

Graph data structures

Node/edge atribution: implementation considerations

Schema and instance representation

Node/edge identification: ObjectsI-Ds vs Values Support for complex relations (hyperedges = n-ary relations)

Query features

Declarative QL: SPARQL, Prolog, Lisp, GraphQL Reasoning: RDF(S), OWL Analysis: Social Networking, graph statistics

Integrity constraints

Most oriented to be schema-less Graph integrity constraints: theoretical interest

Support for essential graph queries

Outline 1. Introduction 2. Current graph databases 3. Comparison of graph database models 4. Conclusions and future work

Conclusions
General balance of graph database models Data structures:
Several types of graph structures Good expressive power for data modeling

Query features:
Restricted to provide APIs Good support for essential graph queries Lack of graph query languages

Integrity constraints:
Basic notions of restrictions Oriented to be schema-less

Future work
Empirical evaluation of graph databases Development of a Benchmark for GDBs Comparison with other database technologies (in particular RDF databases)

Thanks for your attention! Questions?